Allow compilation of ldlt tests if targeting Windows

[mjacobse]

Use Windows-specific _aligned_malloc and _aligned_free for implementing the AlignedAllocator in the ldlt tests on Windows.

I did consider making this completely platform-independent by upping the C-standard from C99 to C11 and just using aligned_alloc. Unfortunately, Microsoft does not support it (see https://learn.microsoft.com/en-us/cpp/standard-library/cstdlib?view=msvc-170#remarks-6) and so MinGW does not seem to provide it either. One could at least remove the posix-specific branch but that did not seem worth the standard change.

I believe another option would be to implement the aligning as in the SimpleAlignedAlloc:

spral/src/ssids/cpu/SimpleAlignedAlloc.hxx

Lines 37 to 52 in b1b7397

	T* allocate(std::size_t n)
	{
	// alloc sufficient space to allow for alignment and storage of base addr
	size_t size = n*sizeof(T) + align;
	void *mem = malloc( size+sizeof(void*) );
	// align it, allowing for base addr beforehand
	void *ptr = reinterpret_cast<void*>(
	reinterpret_cast<char*>(mem) + sizeof(void*)
	);
	if(!std::align(align, n*sizeof(T), ptr, size)) throw std::bad_alloc();
	// store base address to left of returned pointer
	void **vptr = reinterpret_cast<void**>(ptr) - 1;
	*vptr = mem;
	// finally return pointer
	return static_cast<T*>(ptr);
	}

However, this was introduced in fea9ed8 for "testing purposes" and does not seem to be used or tested anywhere at the moment. The only mention of it is commented out:

spral/src/ssids/cpu/NumericSubtree.hxx

Line 35 in b1b7397

//typedef SimpleAlignedAllocator<T> PoolAllocator;

So with those not entirely obvious reinterpret_casts for storing the original allocated pointer this did not seem like the best option to me either.

[jfowkes]

Thanks @mjacobse, I agree that this is the cleanest solution given Microsoft's incomplete C11 support.